Extraction of interesting association rules using genetic algorithms

Abstract

The process of discovering interesting and unexpected rules from large data sets is known as association rule mining. The typical approach is to make strong simplifying assumptions about the form of the rules, and limit the measure of rule quality to simple properties such as support or confidence. Support and confidence limit the level of interestingness of the generated rules. Comprehensibility, interestingness and surprise are metrics that can be used to improve on interestingness. Because these measures have to be used differently as measures of the quality of the rule, they can be considered as different objectives of the association rule mining problem. The association rule mining problem, therefore, can be modelled as multi-objective problem rather than as a single-objective problem. In this paper we present a Pareto−based multi−objective evolutionary algorithm rule mining method based on genetic algorithms. We use confidence, comprehensibility, interestingness, surprise as objectives of the association rule mining problem. Specific mechanisms for mutations and crossover operators together with elitism have been designed to extract interesting rules from a transaction database. Empirical results of experiments carried out indicate high predictive accurracy of the rules generated.